PDA

View Full Version : I would like to have something like this in LW9



Ztreem
11-18-2005, 03:50 AM
I just read that finalrender now supports cinema4d. OK, that wasn't a big surprice, but when I looked at their site and found this render and saw the render time I was amazed! Is it possible we'll see something like this in the new render engine in LW9.

finalrender benchmark (http://www.cebas.com/products/feature.php?UD=10-7888-33-788&PID=53&FID=487)

toonafish
11-18-2005, 03:58 PM
that's very fast !! But Fprime is not bad either, way slower but not as bad as what they used as a comparison. I did a little test with a simular scene. It took Fprime 2.1 about 10 minutes to render this image with Monte Carlo and 2 bounces on a Dell XPS M170 with a Pentium M 780 ( 2.26GHz )

http://www.xs4all.nl/~rponnie/upload/rings.jpg

and about 6 minutes with 1 bounce:

http://www.xs4all.nl/~rponnie/upload/ring_onebounce.jpg

Ztreem
11-18-2005, 05:26 PM
Nice tests! Yeah, Fprime seems quite fast, but with the speed of finalrender( or vray or any other nice renderer out there that not supports LW) you could render 25 frames while fprime do one and you can't even notice the differnce. I think it's about time Newtek starts too implement some new radiosity algorithms, I mean it hasn't change since LW 7 or something. It sounds like they are doing some really big changes on the rendering core for LW9 so why not put in a nice fast radiosity solution that we can play with. :)

Dodgy
11-18-2005, 05:26 PM
This only only took 3mins 45 on LW's own renderer... Which would have been lovely, except I got these really odd patches near the back of the orange ring, if you look carefully. Apart from that, it's gorgeous and smooth, so I can only assume it's some kind of bug... Hopefully one which will get sorted with 9, and since they claim a 2.5 average speed increase on the 9 feature list would make it 1.5 mins.... That is all speculation, but hopefully informed speculation. This was on an Athlon XP 2600+ with 1 gig of ram. It's really odd, gorgeous and smooth at the front of the ring, but that odd blobbing at the back..

Ztreem
11-18-2005, 05:47 PM
Nice! Really good accept for the blotches. If we say that LW 9 renders 2.5 times faster than now it's still 6.5 times slower then finalrender(and that's without blotches)... I don't know what to say other than I want that more, But I don't want to use 3dsmax to get it, I've already ordered my copy of LW9 so it's not about jumping ship or something like that.
I can surely make LW render a picture like that in a couple of seconds but I'll have to do some workarounds. I think it would be nice if we could get some of the goodies all the other 3D appz have without using our old workarounds all the time. That's my thoughts.

connerh
11-18-2005, 08:30 PM
One thing you have to keep in mind, is that Newtek is working on a full program, whereas Cebas is writing a rendering program to be added into programs like LW and 3d Studio Max. If Newtek focused ONLY on the renderer, they probably could get it into high quality/high speed... but I'd rather they develop on all aspects than focus entirely on one.

Also, the workarounds that you're complaining about... every program has those. If you want a faster render, you have to cut corners, granted that on other renderers you may have to cut quite a few less, but people still have to know tricks for each renderer.

Ztreem
11-19-2005, 05:44 AM
Offcourse! I'm with you, I just like to dream.
I really hope they focus on all aspects of LW not only the renderer, but I would be very pleased with their job on the renderer if they implemented something like this. Thats all...

PS. If they do everything in the feature list of 9, I'll will be satisfied... But you always want more, don't you. :)

KillMe
11-19-2005, 09:47 AM
ok done some fprime tests

as it happens on almost the same spec machine dual 248 opteron which i think is 2.2 ghz but i have virus scanners and spyware protection and stuff running aswell as windows media player to thats gotta cancel the little benefit from faster cpu's

first one is 14 secs 1 bounces
second is 14 secs 2 bounces
third is 14 secs and 3 bounces
last is 3mins 3 bounces

spirit_of_stars
11-19-2005, 10:39 AM
Connerh is right... Check the native render of 3DSMax... It not the best. GI rendering is similar to LW renderer... Maybe a little bit more flexible but not powerfull as Vray renderer

toby
11-19-2005, 10:42 AM
Final Render takes advantage of pre-calculated or saved radiosity solutions, I seriously doubt that 14 second render included calculating the whole solution the way LW does.

In other words, LW just needs to be able to save ( and add-to ) the radiosity solutions and it will be in the same ballpark as the other renderers.

Perhaps when NT is done opening up LW's SDK, we'll have access to more renderers...

Captain Obvious
11-20-2005, 11:12 AM
Some things annoy me with that... They compare it to a bi-directional path tracing (BDPT) rendering, and use the result to claim that both BDPT and MLT are unfit for production rendering, and ignore the fact that the whole idea with MLT is that it's much faster than regular BDPT. I'm fairly sure Maxwell, for example, is slower than finalRender, but honestly, it would look a lot better than that after one whole hour. That looks more like a 20 second Maxwell rendering.

toby, if that finalRender image is rendered using a pre-calculated radiosity solution, 14 seconds is pretty ****ed bad, since you could do it in real-time with OpenGL...

Also, this is LW's native renderer, using interpolated radiosity. It took about 170 seconds. Much slower than finalRender, yes, but that was on a 1.33GHz iBook... ;) I'm fairly sure a dual 2GHz PC would do it under a minute. I didn't use shading noise reduction, because I think it reduces the details more than I want to. Yes, there are a few artifacts, but there seems to be some render errors in the finalRender image as well.

Captain Obvious
11-20-2005, 11:24 AM
Bah humbug. Sorry about the file sizes everyone, but JPEG just didn't cut it, so I used PNG.

ThriJ
11-20-2005, 11:42 AM
If they want to maybe Cebas can make a finalRender Stage-2 for LightWave. Just like connerh said NewTek is making a whole program and not just a render engine. I think NewTek is taking the best approach in that they are making is easier for someone else to make a renderer while they improve what they can.

Just read this Quote from the LightWave Development Model:


Additionally, the team completed extraction of the renderer from Layout and shipped it for the first time as a dll/shlib (the team has been somewhat surprised that the appearance of the rendering module as a standalone entity in the Programs folder went completely unremarked by users). This change in the relationship of the renderer to the rest of the application has a number of implications for the future both for NewTek development and for third party opportunities. It will be possible to ship multiple versions of the renderer optimized per platform (processor included), to implement multiple optional renderers that the user can select from, and to allow for third party rendering options. The renderer is now also much easier to work on in order to implement improvements, as the code was previously not contiguous and well-isolated within the Layout core.

toby
11-20-2005, 12:21 PM
toby, if that finalRender image is rendered using a pre-calculated radiosity solution, 14 seconds is pretty ****ed bad, since you could do it in real-time with OpenGL...

Well, pre-calculated is not the same as baked - but it sounds like FR uses a different type of radiosity anyway

The soft shadows coud be explained by the multiple bounces, and it is a little brighter than your example. Can you crank up the bounces to something like 6?

Captain Obvious
11-20-2005, 01:02 PM
Well, pre-calculated is not the same as baked - but it sounds like FR uses a different type of radiosity anyway

The soft shadows coud be explained by the multiple bounces, and it is a little brighter than your example. Can you crank up the bounces to something like 6?
If you don't include the pre-calculation in the total time, does it really matter if it's baked or not? Anyway, yes, I can crank up the bounces to something like 6. Other than totally killing my optimized settings* ;) it doesn't make an awfully big difference. The area right really close to the ring should have a sharp shadow. I'm sure that WinOSi (http://www.winosi.onlinehome.de/) and all other hyper-realistic renderers would agree that the Fprime rendering is technically more accurate.

*two bounces took less than five seconds to render a snapshot of a small area; six bounces took a minute and a half

Here is another example of how finalRender takes too many shortcuts for an example like this to be photo-realistic.

Captain Obvious
11-20-2005, 01:21 PM
More results! By looking at the Blanos benchmarks, a dual 2.0GHz Xeon seems to be more than twice as fast as my 1.33GHz iBook G4. Cebas claimed to have used an old 2GHz PC. If it was dual Opterons, it wouldn't be old, exactly, so I'm assuming they used Xeons. This image took 98 seconds to render, and I think it looks rather good. I should've used more rays per evaluation, though, as it is slightly blotchy. Shading noise reduction was not used. Now, anyway, if we take 98 seconds and divide it by 2.5, we'll get an approximation of how fast this would render on a dual 2GHz Xeon system. 98/2.5 is 39. 39 seconds is still less than half as fast as finalRender, but not too shabby either, what with LW9's improvements.

toby
11-20-2005, 01:30 PM
I'd have to agree that FR's render is probably a lot like Interpolated, a much less accurate estimation. The objects don't look they're really sitting on the plane like yours do, because of those shadows. It's missing what we call the 'contact shadow'.

So it looks like they did the tests in the most complimentary conditions.
That's Marketing for you!

Dodgy
11-20-2005, 01:56 PM
Yup, as with all of these metrics, the people selling the product always pick the best conditions and tweak it :)

Ztreem
11-20-2005, 01:57 PM
Captain O: Ok! that last one is quite good. Now if it only worked with animations...

Captain Obvious
11-20-2005, 02:10 PM
Well, of course they do that. I don't see why we should expect them to show off all the stuff finalRender does poorly. ;) Still, I'm not all that impressed. Their "omg wow fast" image has all sorts of render errors and their area shadows are noisy (http://www.cebas.com/products/images/stage2cinema/ando_fr03_big.jpg).



Captain O: Ok! that last one is quite good. Now if it only worked with animations...
What do you mean? The flickering? Well, with say 17 anti-aliasing passes, the flickering isn't really that bad, I think.

RedBull
11-20-2005, 04:55 PM
I think as mentioned, it's hard for a small company like NT, to compete with the likes of dedicated render makers cebas, who sell plugin renderers for the same price as LW's complete toolset around $700US for FR for C4D.
Especially as Allen, who made our GI and renderer has left NT and already built Modo's renderer. Making it that little bit harder for NT...

The fact that LW9 should allow FPrime to support Nodal shaders,
should make FPrime more attractive for GI.

I would like to see LW9.x with saveable Shadowmaps and GI Calculations....
That would be an easy way to help with some rendertimes in these areas.

trygve
11-20-2005, 05:40 PM
Especially as Allen, who made our GI and renderer has left NT and already built Modo's renderer. Making it that little bit harder for NT...


Well, let's not forget they got Mark Granger on the team in march :bowdown:

"Among Grangerís notable achievements is the development of the Electric Image rendering engine, Camera, which is widely regarded one of the fastest and highest quality renderers on the market. The EI renderer is capable of rendering extremely complex scenes containing many millions of polygons in just a few minutes even when fully anti-aliased at film resolution.

As co-founder of Electric Image, Inc., and the technical visionary behind their leading professional 3D animation system, Granger was responsible for the development of a series of breakthrough technological advances including: invention of a very efficient rasterizing hidden surface algorithm; invention of the motion vector blur algorithm, a first of its kind in the industry capable of producing high quality motion blur streaks in minimal rendering time; invention of a volumetric light ray and smoke engine, another industry first; and the development of an efficient volumetric BSP tree ray tracing engine."

http://www.newtek.com/news/releases/03-09-05b.html

RedBull
11-20-2005, 08:24 PM
I must admit i know very little of EI.... And it's abilities.
So i guess i really can't comment. But fingers crossed! :)
It's an impressive sounding resume alright....

The team NT has assembled is again quite impressive...
So i hope we get some nice results for LW9 and beyond.

Well i'd rather we get faster LW GI (volumetrics too would be nice!)
then have to go and use C4D and FR, for that price...... Oouch
although C4D and Bodypaint, have become quite a good little program, but
considering FPrimes other abilites, i won't be jumping on FR anytime soon.

toby
11-21-2005, 01:29 AM
In the 90's it was one of the best renderers you could get. Fast rendering millions of polys, high quality, and possibly better motion blur than LW has now. Used heavily in Episode 1 ( on a MacIntosh, thank you very much ).

Weepul
11-21-2005, 01:36 AM
The area right really close to the ring should have a sharp shadow. I'm sure that WinOSi (http://www.winosi.onlinehome.de/) and all other hyper-realistic renderers would agree that the Fprime rendering is technically more accurate.

Here is another example of how finalRender takes too many shortcuts for an example like this to be photo-realistic.

Just to play devil's advocate...I mean, render speed's advocate :D...

Here's two points.

1. On TV or the big screen, in motion (animations are where render times need to be the lowest, and "80% in 20% of the time" is most important) do you think it would make that big a difference if the transitions in shadows are exactly as sharp as they need to be? And I daresay their example is both better-looking and faster than anything LightWave's renderer could put out. Maybe, maybe if Interpolated radiosity interpolated its samples in a way that looked smooth, rather than blotchy and error-prone.

2. I haven't used finalRender but I would be surprised if you couldn't change the settings to get a more accurate image with a longer render time. If true, this would show flexibility.

I'm all for accurate stills. I'd be surprised if finalRender couldn't do those. Here, it's showing impressive ability to hold up quality when speed is needed.

To change the topic a bit...anyone else salivating at the examples of finalDOF and finalMotionBlur?

Captain Obvious
11-21-2005, 02:06 AM
1. On TV or the big screen, in motion (animations are where render times need to be the lowest, and "80% in 20% of the time" is most important) do you think it would make that big a difference if the transitions in shadows are exactly as sharp as they need to be?
You have a point. ;) But my point was that their vaunted "as good as a physical light simulation in a fraction of the time" stuff was, well, made up. It's not photo-real. It's quite fast and quite smooth, but not very photo-real.



And I daresay their example is both better-looking and faster than anything LightWave's renderer could put out. Maybe, maybe if Interpolated radiosity interpolated its samples in a way that looked smooth, rather than blotchy and error-prone.
You wouldn't say that if you saw my 17-pass even more optimized rendering. Silky smooth.




2. I haven't used finalRender but I would be surprised if you couldn't change the settings to get a more accurate image with a longer render time. If true, this would show flexibility.
Lightwave allows that as well.

toby
11-21-2005, 02:26 AM
1. On TV or the big screen, in motion (animations are where render times need to be the lowest, and "80% in 20% of the time" is most important) do you think it would make that big a difference if the transitions in shadows are exactly as sharp as they need to be?
In TV, maybe not, unless it was my shot. In film, h*ll yes.


And I daresay their example is both better-looking and faster than anything LightWave's renderer could put out.
Sorry, you're wrong, FR maybe faster, but LW is quite capable of making images every bit as good as FR, and probably with less difficulty.


Maybe, maybe if Interpolated radiosity interpolated its samples in a way that looked smooth, rather than blotchy and error-prone.
It can, it just takes longer.


2. I haven't used finalRender but I would be surprised if you couldn't change the settings to get a more accurate image with a longer render time. If true, this would show flexibility. I'm all for accurate stills. I'd be surprised if finalRender couldn't do those. Here, it's showing impressive ability to hold up quality when speed is needed.
Of course it can do more accurate radiosity. They just don't show us because it doesn't give the impression of hyper-speed.

Speaking of EI, in recent years it shipped with a program that calculated a radiosity solution, the native renderer used it because it couldn't do anything like that. Of course it was very fast, but the light wouldn't change if you moved objects. FR re-uses and modifies this type of solution to save time.

Weepul
11-21-2005, 03:15 AM
Ack, I'm being misintepreted. Or maybe I simply wasn't clear.

The two main thrusts of what I was saying is that:

A. if you're bashing finalRender because the images they showed were not absolutely photorealistic, well, OK - but that's not the point. Both LightWave and finalRender can (likely, in the case of fR) produce absolutely accurate images. Which one would be faster? Dunno. That's not the point. The "upper end" is easy. The "upper end" fast is quite possibly impossible with today's hardware and rendering techniques, but there is development toward that goal (MLT for example).

The point is which one does the "lower end" better-looking and faster.

B. When I said "I daresay their example is both better-looking and faster than anything LightWave's renderer could put out" I didn't mean better-looking independent of speed. I know LW can make an image better-looking than the fR example - with render time measured in minutes. What I was getting at was that within the same ballpark of render time, fR will look much better. More than that, LW would take longer to create a worse result, when the goal is fast renders. No image posted to this thread thus far that was created with the goal of an under-30-seconds rendertime has been close to the fR render time, nor looked as good to my eyes. Captain Obvious' has been the closest so far. How long did your 17-pass silky smooth image take, by the way?

Captain Obvious
11-21-2005, 03:26 AM
A. if you're bashing finalRender because the images they showed were not absolutely photorealistic, well, OK - but that's not the point. Both LightWave and finalRender can (likely, in the case of fR) produce absolutely accurate images. Which one would be faster? Dunno. That's not the point. The "upper end" is easy. The "upper end" fast is quite possibly impossible with today's hardware and rendering techniques, but there is development toward that goal (MLT for example).
Actually, I'm sort of bashing Cebas for claiming total superiority over more accurate algorithms, like MLT, and posted an image that's not very true-to-life and compared it to a path traced image rendered with some horribly unoptimized renderer. Their claims are unfounded, and it annoys me. I just want to put a pin in the balloon that is their inflated egos.


What I was getting at was that within the same ballpark of render time, fR will look much better.
That depends, I think. I greatly prefer some blotchiness or noise over an image like that fR one. The lack of contact shadows is really off-putting in my opinion.



How long did your 17-pass silky smooth image take, by the way?
On a dual 2GHz Xeon, it would have taken about a minute and a half, I think.

Weepul
11-21-2005, 03:56 AM
Actually, I'm sort of bashing Cebas for claiming total superiority over more accurate algorithms, like MLT, and posted an image that's not very true-to-life and compared it to a path traced image rendered with some horribly unoptimized renderer. Their claims are unfounded, and it annoys me. I just want to put a pin in the balloon that is their inflated egos.I can agree with that with the note that their bombastic claims don't negate any objective superiority of their product which may or may not exist.

(Their path tracing example was just ludicrous though.)



That depends, I think. I greatly prefer some blotchiness or noise over an image like that fR one. The lack of contact shadows is really off-putting in my opinion.I'd agree with that too, except I feel the blotchiness or noisiness of a LW render that would take a similar amount of time would be too much for me to accept, whereas I could accept the example fR image - if I was in a situation where I needed 14-second renders.

I did some tests in LW 8.2, giving my dual 1.25 G4 a handicap of about 1.8x (judging from Blanos scores on Tracer Radiosity) so the rendertime target was about 25 seconds. I dumped the results and the scene/object I used here (http://www.stanford.edu/~btaenzer/ringtests/).

By the way, I've only at random been able to get interpolated radiosity to cooperate very well. In the interest of learning, are there any tricks to it? For example, my test scene, I can't even get illumination to hit the ground plane with any settings that would render within the time limit. Other times, it refuses to use smaller sample spacing even when an object is close to a ground plane, or at random will leave whole areas blankly illuminated... Any help you could offer?